Lab 4

Unsupervised classification, also known as clustering, is a machine learning technique used to group data points into clusters based on their similarities. The process of K-means classifcation is first to determine how many clusters you want, then randomly select centriod which will have points assigned to them. Then measure the distance between each point and each centriod and determine which point is in which cluster. Repeat using new centriods until no points can be moved around.

Band 1 is useful for imaging shallow water and tracking fine atmospheric particles like dust and smoke. Band 1 is highly susceptible to atmospheric interference, such as scattering and absorption by gases and aerosols. This interference can distort the spectral signatures of surface features, making it less reliable for classification compared to other bands. Band 1 reflects blues and violets, where light is scattered by dust, smoke, and water particles in the air.

Class 1 = Outer area of map where clip ends

Class 2 = Forest

Class 3 = Water

Class 4 = Forest

Class 5 = Water

Class 6 = Water

Class 7 = Forest

Class 8 = Fields/Agriculture/Golf Courses

Class 9 = Developed Urban

Class 10 = Snow

I would suggest a 10 class unsupervised classification for the most informative results and then combine everything together using the combine tool . This is because 10 classes can sperate many things that the 5 class method missed for exmple brighter surface reflectance does not equal snow. Also 5 classes method does confuse a lot of the urban areas with argicultural land which in the 10 classes is more seperated.

Water – 10 polygons, 50000 pixels

Forest – 10 polygons, 12000 pixels

Low intensity – 11 polygons, 4000 pixels

High intensity – 10 polygons, 6000 pixels

Bare soil – 11 polygons, 500 pixels

Non-forest green veg – 10 polygons – 3000 pixels

I changed a lot of the training data to include more of the spectral reflectance pixels into the right classes. I tried to make more ROI of the water pixels in the Fraser River but when going to the classification it still doesn’t classify the water. These changes affected the quality of my classification by a lot as I got decrease the areas of unclassified pixels and changed pixels into right classes.

The majority analysis changes spurious pixels within a large single class to that class. You enter a kernel size and the center pixel in the kernel will be replaced with the class value that the majority of the pixels in the kernel has. The tool changes the appearance by earasing a couple of outlier classed pixels and cleans up the map in whole. The tool achieve that effect because the outlier pixel is replaced by the what the pixel around it is.

The supervised maximum likelihood classification method does a better job. This is because although there are unclassifed areas in the supervised method, it has overall better classification in terms of classes and areas seperated based on pixels. In the unsupervised method pixels are assigned pixels based on similarity in spectral values. Since unsupervised classification does not rely on class labels, it categorizes all pixels into clusters, even if some clusters may not correspond to meaningful land cover or land use classes. As a result, there are no unclassified pixels in the unsupervised image because all pixels are assigned to clusters, but some clusters may represent mixed classes, or background information. In the supervised classification method, it uses training data to sort based on predefined class labels. The supervised classification algorithm may classify different pixels as unclassified or assign them to a class with low confidence, especially if the spectral signatures of the pixels do not closely resemble any of the training samples. Therefore the supervised classification method may have unclassified classes. With the results from unsupervised classification we could layer that on top of the supervised classification and see which areas that are unclassified in supervised corresponds to which in the unsupervised.